Skip to content

Latest commit

 

History

History
111 lines (86 loc) · 3.52 KB

README.md

File metadata and controls

111 lines (86 loc) · 3.52 KB
Raw

Weight-Change-Prediction-for-Automated-Depression-Diagnosis

This code is for Weight-Change-Prediction-for-Automated-Depression-Diagnosis submitted to ICCAS 2021.

Install

git clone https://github.com/sejong-rcv/Weight-Change-Prediction.git
cd Weight-Change-Prediction

Datasets

We collected ID photos and selfies of celebrities, and generated synthetic data using the PhotoWorks tool. The dataset we released is the 3D landmarks extracted from crawled images.

Synthetic data are processed into five stages for weight change.

  • level_1 : Excessive Weight Loss
  • level_2 : Little Weight loss
  • level_3 : Original Image
  • level_4 : Little Weight Gain (eg. Swelling)
  • level_5 : Excessive Weight Gain
./Weight-Change-Prediction/
├── Synthetic_Dataset/
│   ├── 001_1_landmark.json # person ID 001's level(1 & 3) landmark
│   ├── 001_2_landmark.json # person ID 001's level(2 & 3) landmark
│   ... 
│   └── 500_5_landmark.json # person ID 500's level(5 & 3) landmark
│
├── Real_Dataset/
│   ├── real_002_landmark.json
│   ├── real_001_landmark.json
│    ... 
│   ├── refine_012_landmark.json
│   └── refine_013_landmark.json
│
├── train_list_final.txt          # train file
└── test_list_final.txt           # test file

The weight change level was randomly sampled and used for training, so it provides a path to the train and test files as a txt file.

json files are configured as follows:

[
  { "level"     # Synthetic Image's Level for comparison with the original
    "landmark"  # Landmark of Synthetic Image (468, 3)
    "label"     # 1 or 0 : When there's a weight change, label is 1
  }, 
  {
    "level"     # Original Image Level
    "landmark"  # Landmark of Original Image (468, 3)
    "label"     # the same level as the Synthetic image in same file
  }
]

Implementation

We provide colab tutorial, and full guidance. There are also tutorials for training the model, extracting features from landmarks, and visualization for 3D Face Landmark. (you can check the colab tutorial code in ipynb file)

Set the data directory and pretrained model path –

DATA_DIR = "/Weight-Change-Prediction" # Data Directory
load_model = None # if you want to use pretrained model, insert the .pkl file path

To train and evaluate the model, run this file –

python train_eval.py

Performance

Model Accuracy
Light Gradient Boosting (LightGBM) 93.00
K Nearest Neighbor (KNN) 92.75
Random Forest (RF) 92.25
Linear Discriminant Analysis (LDA) 89.25
Decision Tree (DT) 88.00
LogisticRegression (LR) 87.00
Quadratic Discriminant Analysis (QDA) 69.00
Depth Accuracy
30cm 0.7
30cm 0.7
45cm 0.9

References

Weight-Change-Prediction-for-Automated-Depression-Diagnosis (ICCAS)

Juyoung Hong, Jeongmin Shin, Yujin Hwang, Jeongmin Lee, and Yukyung Choi

@inproceedings{weight_change_prediction,
  author = {Juyoung Hong, Jeongmin Shin, Yujin Hwang, Jeongmin Lee, and Yukyung Choi},
  title = {Weight-Change-Prediction-for-Automated-Depression-Diagnosis},
  booktitle = {International Conference of Contral, Automation and Systems(ICCAS)},
  year = {2021},
}